RPI3 swap experiments (r338342 with vm.pageout_oom_seq="1024" and 6 GB swap)

[Mark Millard]

On 2018-Sep-6, at 8:58 AM, bob prohaska <fbsd at www.zefox.net> wrote:

> On Thu, Sep 06, 2018 at 12:08:09AM -0700, Mark Millard wrote:
>> On 2018-Sep-5, at 10:15 PM, bob prohaska <fbsd at www.zefox.net> wrote:
>> 
>>>> . . .
>>> There were two buildworld tests run with 6 GB of swap, the first without
>>> TRIM being turned on and the second with TRIM turned on. The second run
>>> too an hour longer, with TRIM being on the only difference. 
>> 
>> buildworld did not take an hour longer for one vs. the other
>> based on the timestamps in the log files:
>> 
>> trim off:
>> 
>>>>> World build started on Sun Sep  2 20:28:12 PDT 2018
>> . . .
>>>>> World build completed on Mon Sep  3 21:35:47 PDT 2018
>> 
>> So somewhat over 25 hours 7 minutes.
>> 
>> trim on:
>> 
>>>>> World build started on Tue Sep  4 00:02:36 PDT 2018
>> . . .
>>>>> World build completed on Wed Sep  5 01:12:47 PDT 2018
>> 
>> So somewhat over 25 hours 10 minutes.
>> 
>> I get an under 5 minute difference from those timestamps.
>> 
>> 
> 
> You are correct, the mistake is mine. Thanks for catching it!
> 
>>>> . . .
>>> Near as I can tell there are no non-zero values for d/s, which if it's tied
>>> to TRIM is reasonable for all but microSD, which did have TRIM enabled. Since
>>> microSD wasn't particularly busy, apart from swap, that too is unsurprising.
>> 
>> http://www.zefox.net/~fbsd/rpi3/swaptests/r338342/3gbsd_3gbusb/trim_on/writedelay.sort
>> shows ms/d and its matching kBps with the below non-zero figures:
>> 
>> . . .
>>    0      0      0      0    0.0      0      6   16.2      0     10    0.1    0.3  mmcsd0
>>    0      0      0      0    0.0      0      6   16.3      0     10    0.2    0.3  mmcsd0s2
>> . . .
>>    0      0      0      0    0.0      0      3   99.2      0      4   40.7    1.3  mmcsd0s2a
>>    0      0      0      0    0.0      0      3   99.3      0      4   40.8    1.3  ufs/rootfs
>> . . .
>>    0      0      0      0    0.0      0     10  147.6      0      6   12.0    4.5  mmcsd0s2a
>> . . .
>>    0      2      1      8    1.8      0      7  211.0      0     10  114.5    5.8  mmcsd0
>>    0      2      1      8    1.8      0      7  211.1      0     10  114.5    5.8  mmcsd0s2
>>    0      1      0      0    0.0      0      7  211.2      0     10  114.6    5.6  ufs/rootfs
>> . . .
>> 
>> Those are all I found. I may have missed some.
>> 
> 
> Again, I was mistaken, misreading your original question. 

I looked at examples of using 1 device for swap and only 2 GiByte of
swap ( still -r338342 ):

For 2gbsd (2 GiBytes on the sdcard, two 1 GiByte partitions used):

World build started on Thu Aug 30 09:44:09 PDT 2018
. . .
World build completed on Fri Aug 31 07:56:20 PDT 2018

So: somewhat over 22 hour and 12 minutes.

vs.

For 2gbusb (2 GiBytes on the usb device, two 1 GiByte partitions used):

World build started on Fri Aug 31 10:27:17 PDT 2018
. . .
World build completed on Sat Sep  1 12:49:19 PDT 2018

So: somewhat over 26 hours 22 minutes.

So a difference of over 4 hours.


It appears to me that for the devices that you are currently using,
you are better off for elapsed time having the swap only on the
sdcard.

It may be evidence that the split of heavivly in  use UFS and the
swap partition across the channels/devices is useful. Or it could
just be that the sdcard is faster. I can not tell from the cases
available.

If you had a sufficiently large sdcard of similar performance,
and ran a test that avoided the usb device completely, that would
give a clue (still -r338342 or otherwise matching contexts for the
comparison with a split context).


===
Mark Millard
marklmi at yahoo.com
( dsl-only.net went
away in early 2018-Mar)